The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and implement instructions from external plan files and repository configurations, creating a vulnerability surface where malicious instructions in a plan could lead the agent to perform unauthorized actions.
Ingestion points: Implementation plans (e.g., *-plan-*.md), repository instructions, workspace configuration files (.agent-work/workspace.json), and upstream ticket artifacts.
Boundary markers: The skill utilizes a structured 'phase ledger' to track state and relies on specific reference files (references/phase-handoffs.md) to manage transitions and maintain context separation.
Capability inventory: The agent performs file system inspection, executes repository management commands (Git), and runs 'automated criteria' which can include arbitrary shell scripts or testing tools. It also delegates implementation tasks to subagents.
Sanitization: The skill relies on mandatory human-in-the-loop gates after each implementation phase to review changes and provide authorization before commits are created.
[COMMAND_EXECUTION]: The skill performs shell-based operations to interact with the Git version control system and execute verification suites.
Evidence: The skill instructions explicitly utilize ls -La, git status, git diff, and git commit as part of the implementation and verification workflow defined in SKILL.md and references/verification-and-checkpoints.md.

implement-plan